Bounded Dirichlet series
2.2 Vertical limit functions
We will now introduce the concept of a vertical limit function, which will be of great impor-tance in the study of composition operators onH 2. In order to do so, we need the notion of a multiplicative character. We say that χ:N→T is a multiplicative character if it satisfies χ(mn) =χ(m)χ(n) for all positive integersmandn. We denote the set of all such characters by M. It is convenient to identify the set M with T∞, the infinite dimensional Cartesian product of T. This identification is done as follows. Take a point z = (z1, z2, ...) ∈ T∞ and let χ(pj) = zj, where pj denotes the j-th prime. The character χ is now defined for every prime number and we extend it to all of N by letting χ(n) =χ(pr11)· · ·χ(prmm), where n = pr11 · · · prmm. Note that this definition of χ forces it to be multiplicative. Moreover, T∞ is a compact group under point-wise multiplication. It therefore exists a unique Haar measure on T∞, which happens to coincide with the infinite product measure generated by the normalized Lebesgue measure on T.
Consider now the function f(s) = P∞
n=1ann−s ∈ D. For any characterχ∈ M we define the function fχ by
fχ(s) =
∞
X
n=1
anχ(n)n−s. (2.2)
We call this a vertical limit function. To understand the reasoning behind this name we look at the character χ(n) = n−it. This obviously defines a multiplicative character. Also, for f(s) =P∞
n=1ann−s we havefχ(s) =f(s+iτ) =:fτ(s). That is, fχ is a vertical translation of f. The interesting thing is, as we shall see shortly, that every vertical limit function fχ
corresponds to the limit of a sequence of vertical translations of f. We start out by proving a famous theorem by Kronecker, and we will give an analytic proof due to Bohr [3].
Definition 2.6. Let a1, ..., an be a set of real numbers. We say that the numbers are Q -linearly independent if
n
X
j=1
cjaj = 0, with c1, ..., cn ∈Z, only when cj = 0 for 1≤j ≤n.
One typical example of such a set is the following. If p1, ..., pj is a set ofj unique primes, then the real numbers logp1, ...,logpj are Q-linearly independent. This follows from the relation
j
X
i=1
cilogpi = log Y
1≤i≤j
pcii, and the fundamental theorem of arithmetic.
Theorem 2.7 (Kronecker’s Theorem). Suppose ξ1, ..., ξk are Q-linearly independent real numbers. Let α1, ..., αk be arbitrary real numbers and let > 0. Then there exists integers n1, ..., nk and a real number t satisfying
|tξj −αj −nj|< , j= 1,2, ..., k.
Proof. We want to carry out an analytic proof of this result and we will rely on the following fact. The exponential function e2πix is equal to 1 if and only if xis an integer. Consider the function to an integer for all m. In this language, Kronecker’s Theorem states that we can get |f(t)|
arbitrarily close to k+ 1 by choosing t sufficiently large.
Let F denote the function
F(x1, .., xk) = 1 +x1+· · ·+xk.
If we now raise the functions f and F to their p-th power, for some integer p, and use the multinomial theorem, they take the form
(f(t))p =X
bνeiβνt, (F(x1, ..., xk))p =X
bλ1,...,λkxλ11 · · ·xλkk.
Note that the βν must all be different due to the linear independence of the ξi. The p-th power of the functions f and F must therefore have the same number of terms, and the absolute value of the coefficients bν and bλ1,...,λk coincide. This yields
X|bν|=X
bλ1,...,λk = (F(1, ...,1))p = (k+ 1)p. For any fixed ν =ν0 we have the following identity,
Tlim→∞ then every coefficient bν of (f(t))p satisfies
|bν|=
Askincreases by 1, the number of terms in the polynomial development of (1 +x1+...+xk)p is multiplied with at most p+ 1. This means that the total number of terms must be less than (p+ 1)k. We therefore have
X|bν|<(p+ 1)kCp, which implies that
(p+ 1)kCp
P|bν| = (p+ 1)kCp (k+ 1)p >1 for all p. However, since C < k+ 1 we have that
(p+ 1)kCp (k+ 1)p →0,
as p → ∞. Therefore, our assumption that |f(t)| is less than C for every t ∈ R leads to a contradiction. Consequently, the proof is complete.
The result below was originally presented in [10], and we will follow a proof from [13].
Theorem 2.8. Suppose that f(s) = P
n=1ann−s, with σu(f) = θ. Then the vertical limit functions fχ are precisely the limits of some sequence {fτN} of vertical translations of f, in the half-plane Cθ. The sequence {fτN} converges uniformly on Cθ+δ, for every δ > 0.
Moreover, we have σu(fχ) = θ.
Proof. Assume that we have a sequence of vertical translations {fτN}N≥1 converging to an analytic function on Cθ. It is clear that the function defined as ξτN(n) := n−iτN must converge to some limit functionξ(n), as N → ∞. We readily see that |ξ(n)|= 1 and that ξ is completely multiplicative. Hence, there exists a character χ ∈ M such that χ(n) =ξ(n) for alln. Letϑ > θ. We know thatf converges uniformly on Cϑ. We can then take the limit as N approaches infinity of the translations fτN, and pass the limit inside the summation.
This will, in conjunction with the argument above, ensure that fτN converges to fχ on Cϑ. Since ϑ was arbitrary, we actually have convergence on all of Cθ and alsoσu(fχ) = θ.
On the other hand, we want to show that for any characterχ∈ Mthere exists a sequence of real numbers {τN}N≥1 such that {fτN}N≥1 converges to fχ. For this purpose, we recall Kronecker’s Theorem (Theorem 2.7). In the language of vertical limit functions the theorem says that there exists τ ∈R such that
|χ(n)−n−iτ|< ε, n= 1, ..., N,
for anyε >0 and positive integer N. Then for allN we should be able to findτN such that
|χ(n)−n−iτN|<1/N n= 1, ..., N.
As N approach infinity we see thatn−iτN →χ(n), which is what we wanted to prove.
Since all characters χ has modulus 1 it is also clear that σa(fχ) = σa(f). We will now establish a product formula for vertical limit functions. The lemma is a slight improvement of a result from [13].
Lemma 2.9. Assume we have a characterχ∈ M and two convergent Dirichlet seriesf and g. Then the product formula (f g)χ =fχgχ holds in Cθ whenever σu(f)≤θ and σu(g)≤θ.
Proof. In a remote half-plane where both f and g converges absolutely we can write the product f(s)g(s) = (P∞
n=1ann−s) (P∞
m=1bmm−s) as f(s)g(s) =
∞
X
n=1
cnn−s, where
cn= X
ij=n
aibj. Then
(f g)χ=
∞
X
n=1
cnχ(n)n−s. On the other hand we have
fχgχ=
∞
X
n=1
dnn−s, where
dn= X
ij=n
aibjχ(i)χ(j).
But χ is a completely multiplicative character, so dn =χ(n)X
ij=n
aibj =χ(n)cn.
We see that (f g)χ =fχgχin a half-plane wheref andg converges absolutely. Sinceσu(f)≤θ andσu(g)≤θ, the two functions are bounded onCθ. The product of two bounded functions are again bounded, so by Bohr’s theorem it follows that σu(f g) ≤ θ. The product formula therefore holds in all ofCθ.
The next result tells us that the image of a function is invariant under vertical translations.
We follow a proof from [5].
Lemma 2.10. Let ψ : Cθ → Cν be analytic, with ψ(s) = P∞
n=1cnn−s. Then ψ(Cθ) = ψχ(Cθ).
Proof. We observe thatψ(+∞) = ψχ(∞) =c1, so the result is clear for constantψ. Assume now that ψ is non-constant. For a point w∈ Cθ we find a closed disc K which contains w in its interior and satisfies
M = inf
s∈∂K|ψχ(s)−ψχ(w)|>0.
By Theorem 2.8 there exists a sequence of vertical translation of ψ converging to ψχ on K.
Let τk be a sequence such that ψ(s+iτk) → ψχ(s). By choosing k large enough we can ensure that
|ψ(s+iτk)−ψχ|< M.
We add and subtract ψχ(w) on the left-hand side, so that
|ψ(s+iτk)−ψχ(w)−(ψχ(s)−ψχ(w))|< M, valid for s ∈K. Denote by f and g the functions
f(s) = ψχ(s)−ψχ(w)
g(s) = ψ(s+iτk)−ψχ(w)−(ψχ(s)−ψχ(w)).
The function f is clearly zero at s=w. Rouch´e’s theorem tells us that f and f+g has the same number so zeros in K. We have
(f +g)(s) =ψ(s+iτk)−ψχ(w),
so there must be a value of s inK makingψ(s+iτk) =ψχ(w), and the proof is done.